Optimal Transmit Power in Wireless Sensor Networks - 2006

Optimal Transmit Power in
Wireless Sensor Networks

Sooksan Panichpapiboon, Student Member, IEEE, Gianluigi Ferrari, Member, IEEE, and
Ozan K. Tonguz, Member, IEEE
AbstractPower conservation is one of the most important issues in wireless ad hoc and sensor networks, where nodes are likely to
rely on limited battery power. Transmitting at unnecessarily high power not only reduces the lifetime of the nodes and the network, but
also introduces excessive interference. It is in the network designers best interest to have each node transmit at the lowest possible
power while preserving network connectivity. In this paper, we investigate the optimal common transmit power, defined as the
minimum transmit power used by all nodes necessary to guarantee network connectivity. This is desirable in sensor networks where
nodes are relatively simple and it is difficult to modify the transmit power after deployment. The optimal transmit power derived in this
paper is subject to the specific routing and medium access control (MAC) protocols considered; however, the approach can be
extended to other routing and MAC protocols as well. In deriving the optimal transmit power, we distinguish ourselves from a
conventional graph-theoretic approach by taking realistic physical layer characteristics into consideration. In fact, connectivity in this
paper is defined in terms of a quality of service (QoS) constraint given by the maximum tolerable bit error rate (BER) at the end of a
multihop route with an average number of hops.
Index TermsAd hoc wireless networks, sensor networks, power control, connectivity.
1 INTRODUCTION
I
N an ad hoc wireless network, where nodes are likely to
operate on limited battery life, power conservation is an
important issue. Conserving power prolongs the lifetime of
a node and also the lifetime of the network as a whole. In
addition, transmitting at low power reduces the amount of
excessive interference. The fundamental question which
naturally arises is: What is the optimal transmit power to be
used? This is the fundamental question that we try to
answer in this paper. Obviously, a suitable criterion of
optimality has to be introduced.
One of the goals of forming a network is to have network
connectivitythat is, each node should be able to commu-
nicate with any of the other nodes, possibly via multiple
hops. The connectivity level of an ad hoc wireless network
depends on the transmit power of the nodes. If the transmit
power is too small, the network might be disconnected (i.e.,
there may be multiple disconnected clusters of nodes
instead of a single overall connected network). However,
as mentioned earlier, transmitting at excessively high
power is inefficient because of the mutual interference in
the shared radio channel and the limited battery lifetime.
Thus, it is intuitively clear that the optimal transmit power
is the minimum power sufficient to guarantee network
connectivity [1], [2], [3].
Ideally, the transmit power of a node should be adjusted
on a link-by-link basis to achieve the maximum possible
power savings [2], [3], [4], [5], [6]. Nonetheless, due to the
absence of a central controller in a pure ad hoc network
with flat architecture, performing power control on a link-
by-link basis is a complicated and cumbersome task. A
simpler solution, which is more viable for implementation,
is to have all the nodes use a common transmit power. This is
desirable in sensor networks where nodes are relatively
simple and it is difficult to modify the transmit power after
deployment. In addition, the performance difference, in
terms of traffic carrying capacity, between adjusting the
power locally and employing a common transmit power is
small, especially when the number of nodes is large [1].
In this paper, we investigate the optimal transmit power
for an ad hoc wireless networking scenario where all nodes
use a common transmit power. Although the optimal
common transmit power derived in this paper is subject
to the routing and the medium access control (MAC)
protocol considered, the approach can be extended to other
routing and MAC protocols as well. Other studies which
consider common transmit power exist [1], [7], [8], [9]. In [7]
and [8], the minimal transmission range at which a network
is connected with high probability is studied. In [9], the
authors investigate the minimal common transmit power
sufficient to preserve network connectivity. These works,
however, follow a graph-theoretic approach which only
takes into account the distances between nodes. More
specifically, the authors consider that two neighboring
nodes can communicate if they are within the communica-
tion range of each other and two nodes that are not
neighbors can communicate if there is a multihop path
connecting them. It is important to point out that, although
there may be a path connecting two nodes, communication
1432 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 5, NO. 10, OCTOBER 2006
. S. Panichpapiboon and O.K. Tonguz are with the Electrical and Computer
Engineering Department, Carnegie Mellon University, 5000 Forbes Ave.,
Pittsburgh, PA 15213-3890.
E-mail: sooksan@cmu.edu, tonguz@ece.cmu.edu.
. G. Ferrari is with the Dipartimento di Ingegneria dellInformazione,
University of Parma, Viale G.P. Usberti, 181/A, I-43100 Parma, Italy.
E-mail: gianluigi.ferrari@unipr.it.
Manuscript received 11 Feb. 2005; revised 22 Aug 2005; accepted 5 Nov.
2005; published online 16 Aug. 2006.
For information on obtaining reprints of this article, please send e-mail to:
tmc@computer.org, and reference IEEECS Log Number TMC-0029-0205.
1536-1233/06/$20.00 2006 IEEE Published by the IEEE CS, CASS, ComSoc, IES, & SPS
between them may not be possible as the quality of service
(QoS) in terms of tolerable bit error rate (BER) at the end of
a multihop route may not be satisfied. We discuss this in
more detail in Section 3. As opposed to the conventional
graph-theoretic approach, in this paper, the optimal
transmit power sufficient to maintain network connectivity
is found according to a physical layer-oriented QoS
constraint given by the maximum tolerable BER at the
end of a multihop route with an average number of hops.
In this paper, we evaluate the optimal transmit power
both analytically (in the case of regular topology) and via
simulations (in the case of random topology). Moreover, we
investigate the interrelation between optimal transmit
power, data rate, and node spatial density. In addition to
our earlier work [10], this paper also investigates 1) the
impact of different propagation pathloss exponents (on
different links of a multihop route) on the performance of a
common transmit power control scheme and 2) the
interrelation between transmit power, connectivity, and
network longevity. Furthermore, in this paper, we provide a
rigorous analysis of exact interference power and BER using
a detection theory approach, as opposed to simply assum-
ing that the interference on each link is the average power
as considered in [10]. We also validate our analysis with
simulation.
The rest of this paper is organized as follows: In Section 2,
we describe the model and the assumptions that will be
used in the derivation of the optimal transmit power. In
Section 3, we define network connectivity. We introduce a
communication-theoretic model for evaluating route BER in
Section 4. The minimum transmit power sufficient to
maintain network connectivity in both regular and random
topologies is analyzed in Section 5. The performance in
terms of node/network lifetime and effective transport
capacity is evaluated in Section 6. Numerical results, along
with their implications, are presented in Section 7. Finally,
we provide a discussion of related work and conclusions in
Section 8 and Section 9, respectively.
2 MODEL AND ASSUMPTIONS
In this section, we describe the basic ad hoc wireless
network communication model and the basic assumptions
considered in the this paper.
2.1 Network Topology
Throughout the paper, we consider a scenario where
` nodes are distributed over a surface with finite area .
The node spatial density is defined as the number of nodes
per unit area and is denoted as ,
s

4
`,. To avoid edge
effects, we assume the network surface to be the surface of a
torus with length 21 on each edge, as shown in Fig. 1.
However, the analytical technique presented in this paper
can be applied to other types of surfaces as well. In a real
scenario, the performance predicted by our analysis may
not be extremely precise for nodes on the edge of the
network surface. In this case, a more precise performance
evaluation may be obtained via simulations. Nonetheless,
the results presented in this paper provide a representative
description of a realistic network behavior.
In addition to a simple scenario with square grid
network topology considered in [10], in this paper, we also
consider a realistic scenario with two-dimensional Poisson node
distribution. In a network with square grid topology, shown
in Fig. 1a, each node has four nearest neighbors at a fixed
distance. In contrast, the positions of nodes in a network
with a two-dimensional Poisson topology are random and
independent of each other, as shown in Fig. 1b. The
probability mass function (pmf) of the number of nodes `
o
over a surface of area o in the case with two-dimensional
Poisson topology is given by
Pr`
o
,
,
s
o
,
,!
c
,so
, 0. 1. 2. . . . . 1
where, in this case, ,
s
corresponds to the average number of
nodes per unit area or the average node spatial density.
Considering the same value of ,
s
for both types of networks
with regular and random topologies makes the comparison
between them fair and meaningful.
In this paper, we only consider ad hoc wireless networks
with stationary nodes. Examples of such networks are sensor
networks [11] and wireless mesh networks [12]. The
extension to a scenario where nodes are mobile can be done
following the approach proposed in [13].
2.2 Routing
We assume a simple routing strategy such that a packet is
relayed hop-by-hop, through a sequence of nearest neigh-
boring nodes, until it reaches the destination. In addition,
we assume that a source node discovers a route prior to
data transmission [14]. Discovery of a multihop route from
a source to a destination is a crucial phase in a wireless
networking scenario with flat architecture. The focus of this
paper, however, is on the characterization of the steady
state behavior of on-going peer-to-peer multihop commu-
nications. Therefore, we will assume that a route between
source and destination exists. We discuss routing in
networks with regular and random topologies in the
following sections.
2.2.1 Square Grid Topology
Due to the regularity of this topology, the distance to the
nearest neighbor, denoted by i
link
, is fixed, and a route is
PANICHPAPIBOON ET AL.: OPTIMAL TRANSMIT POWER IN WIRELESS SENSOR NETWORKS 1433
Fig. 1. Possible topologies: (a) regular and (b) random. In each case,
examples of multihop routes are shown.
constituted by a sequence of hops with equal length. The
distance i
link
can be computed as follows: One can first
observe that constructing a square lattice of ` nodes over a
surface of a torus with area is equivalent to fitting `
small square tiles of area i
2
link
into a large square of area .
Hence, it must hold that `i
2
link
and, therefore, the
distance to the nearest neighbor can be written as
i
link

`
r

1
,
s
p . 2
2.2.2 Two-Dimensional Poisson Topology
In the case of random topology, we still consider a routing
scheme where each intermediate node in a multihop route
relays the packets to its nearest neighbor in the direction of
the destination. In particular, we assume that an intermedi-
ate node in the route selects the nearest node within a sector
of angle 0 toward the direction of the destination as the next
hop [15]. An example of a multihop route constructed in this
way is shown in Fig. 2. In this case, a route can be visualized
as a deviation from the straight line between source and
destination, referred to as reference path. Unlike in a scenario
with grid topology, in a network with two-dimensional
Poisson topology, the distance from a node to its nearest
neighbor is not a constant. Let \ be a random variable
denoting the distance to the nearest neighbor in a two-
dimensional Poisson node distribution. It can be shown that,
keeping the node spatial density fixed, for large ` (i.e., as
` !1), the CDF of the distance to the nearest neighbor in a
torus is
1
\
n
0 n < 0
1 c
,
s
n
2
0 n < 1
1 c
,s
1 n <
2
p
1
1 n !
2
p
1.
8
>
>
<
>
>
:
3
where
4
4n
2
4

41
n
2
1
2
p
h i
and
4
cos
1 1
n

.
Note that the total surface area of a torus with length 21 on
each edge is equal to 41
2
. Hence, the node spatial density in
this case is ,
s
`,41
2
. Substituting this value back in (3)
and computing 1
\
1, it follows that Pr\ 1, i.e.,
1
\
1, is almost 1 if ` ! 10. In other words, \ is almost
surely lower than 1. Consequently, for ` )10 (which will
be true for all scenarios considered in this paper), we can
neglect the case where \ 1 without significant impact
on the accuracy of the analytical results.
Using the fact that \ 1 with very high probability and
generalizing the CDF given in (3) to a scenario where a node
looks for a neighboring node within a sector of angle 0, the
CDF of \ can be written as
1
0
\
n
1 n 1
1 c
,
s
0n
2
2
0 n 1
0 otherwise.
8
<
:
4
More details on the derivation can be found in [16].
2.3 Medium Access Control Protocol
In this paper, we consider a simple reservation-based MAC
protocol introduced in [17] and defined as reserve-and-go
(RESGO).
1
In this protocol, a source node first reserves
intermediate nodes on a route for relaying its packets to the
destinationcharacterization of this phase is beyond the
scope of this paper. A transmission can begin after a route
is discovered and reserved. The main idea of the protocol is
that a source node or a relay node generates an exponential
random backoff time before it transmits or relays each
packet. After the random backoff time expires, a node can
start transmitting a packet. The random backoff time helps
reduce interference among nodes in the same route and
also among nodes in different routes. Throughout this
paper, we assume that the random backoff time is
exponential with mean 1,`
t
. In other words, given that a
node has packets to send, packets are transmitted with rate
`
t
(dimension: [pck/s]). Note that this is generally different
from the traffic generation rate, `
g
.
3 CONNECTIVITY
As discussed earlier, the optimal common transmit power is
the minimum power sufficient to preserve network con-
nectivity. In this section, we formalize the definition of
network connectivity. Conceptually, an ad hoc wireless
network is often viewed as a graph, where vertices
represent the nodes and edges represent the links connect-
ing neighboring nodes. From a graph-theoretical perspec-
tive, a network is connected if there is a path (possibly
multihop) connecting any node to any other node in the
network. However, using this notion of connectivity for an
ad hoc wireless network, where a communication channel is
error-prone, can be misleading. Since the wireless links are
susceptible to errors, the QoS in terms of route BER
deteriorates as the number of hops in a route increases.
Consequently, the performance may be unacceptable,
although there is a sequence of links to the destination.
In order to take the physical layer characteristics into
account, in this paper, we consider network connectivity
from a communication-theoretic viewpoint. In particular, a
network is said to be connected if any source node can
communicate with a BER lower than a prescribed value
BER
th
to a destination node placed at the end of a multihop
Fig. 2. Possible multihop route in a random topology. \ corresponds to
the distance to the nearest neighbor, corresponds the projection
angle, and 0 corresponds to the angle which a node looks for a neighbor
in the direction of the destination.
1. This MAC protocol was erroneously referred to in [17] as the Aloha
MAC protocol for its resemblance to the classical Aloha MAC protocol [18].
However, there are significant differences which make the proposed
protocol different from the classical Aloha MAC protocol: 1) multihop route
reservation and 2) no use of retransmission techniques.
route with an average number of hops [19]. To be conserva-
tive, in most of this paper, we consider an ideal worst-case
scenario where an information bit is relayed on each link of a
route toward a destination without retransmissions. How-
ever, it will be shown that use of retransmission techniques
lowers the BER and thus decreases the minimum power
required to maintain the network connectivity. We will
discuss this in more detail in Section 7.1.
In addition, note that this notion of connectivity
corresponds to requiring that, on average, a communication
between a source and a destination can be guaranteed with
a desired quality. However, it does not guarantee that a
source can communicate with every node in the network
with this QoS. A more stringent connectivity requirement,
such that a source can communicate with every node in the
network with the desired QoS, can also be enforced. The
approach proposed in this paper can be straightforwardly
extended by considering the BER at the end of a multihop
route with the maximum possible number of hops. In the next
section, we derive a simple analytical expression for the
average number of hops in the cases with grid and random
topologies.
3.1 Square Grid Topology
Due to the spatial invariance on a torus, we can assume
without any loss of generality that a source node is at the
center of the network (see Fig. 3). If a destination node is
selected at random, the minimum number of hops to reach
the destination can range from 1 to 2i
max
, where i
max
is the
maximum tier order. In other words, it takes 1 hop to reach
a destination which is a neighbor of a source node in Tier 1,
and it takes 2i
max
hops to reach the farthest node from the
center in Tier i
max
. The average number of hops can be
obtained by counting the number of hops on a route from
the source to each destination node and finding the average
value. Assuming that each destination is equally likely, the
average number of hops on a route can be written as
i
grid

1
` 1
4
X
i
max
i1
i 4
X
i
max
i1
2i 8
X
i
max
i1
X
i1
,1
i ,
" #
. 5
The first summation term in (5) corresponds to the number
of hops it takes to reach any of the four nodes in alignment
with the source at the center of the network in all possible
tiers, the second summation corresponds to the number of
hops to reach nodes on the four corners of each tier, and,
finally, the third summation corresponds to the number of
hops to reach the other nodes in each tier. With straightfor-
ward algebra, (5) can be simplified to
i
grid

2
` 1
2i
3
max
3i
2
max
i
max

. 6
Since i
max

`
p
,2, when the number of nodes is suffi-
ciently large, from (6), one obtains:
i
grid

`
p
2

1
`
p
3
2
`
p
2
. 7
3.2 Two-Dimensional Poisson Topology
In this case, we define an average route as a route between a
source and a destination separated by an average Euclidean
distance between two randomly chosen points on a torus.
We now propose an approach for computing the average
path length. Let 7 be the random variable denoting the
distance between a source and a destination. It can be
shown that the pdf of 7 is [15]
)
7
:
:
21
2
0 : < 1
:
21
2
2: cos
1
1,:
1
2
1 : <
2
p
1.
(
8
Based on (8), the following average distance between a
source and a destination node can be obtained:
7
Z

2
p
1
0
:)
7
:d:
1
3
2
p
ln1
2
p

h i
. 9
Intuitively, the average number of hops in a route must
be proportional to the average hop length, i.e., the average
distance between two neighbors. For instance, one can
expect that a route will consist of many hops if each hop
length is short. In our analysis, we assume that each hop
deviates, with respect to the reference path, by an angle ,
where is uniformly distributed in the interval 0,2. 0,2
the angles and 0 are shown in the example considered in
Fig. 2. By projecting each hop onto the reference path, the
average number of hops between a source and a destination
can be approximated as follows:
2
i
rand

7
E\ cos
. 10
where E\ cos is the average projected hop length. In
general, \ and are not independent. However, as
discussed in Section 2.1, since it is highly likely that
\ 1, in this case, it can be shown that \ and are
basically independent. Therefore, it follows that
E\ cos E\Ecos .
The expected values of \ and cos can be straightfor-
wardly obtained fromthe pdfs of \ and . In fact, assuming
Fig. 3. Tier structure of a network with square grid topology.
2. To be precise, in the derivation of i
rand
in (10), one should consider
E\ cos ,0 at the denominator on the right-hand side. However, in the
remainder of this paper, we assume that 0 is given and all the expected
values have to be interpreted as conditional expectations. In general, the
angle 0 is likely to be known. For example, in a routing protocol where a
node does not select the next relay node that is in the opposite direction of
the destination, the angle 0 is .
that the network area is large (i.e., 1 is large), it can be
shown that E\
,2,
s
0
p
and Ecos 2,0 sin0,2
[15]. Finally, using (9), (10), and the simplified expression for
E\ cos , it can be shown that
i
rand

`
p
0
3
p
2
p
ln 1
2
p
6
2
p
sin
0
2
. 11
where we have used the fact that ,
s
`,.
4 BER AT THE END OF A MULTIHOP ROUTE
Since the network connectivity is defined in terms of BER
quality at the end of a multihop route, in this section, we
present an approach to evaluate the route BER.
4.1 Square Grid Topology
4.1.1 BER with Exact Interference Analysis
In this section, we analyze the link BER and the route BER
using a detection-theoretic approach. Generally, the re-
ceived signal observed at the receiver is the sum of three
components: 1) the intended signal from a transmitter,
2) the interfering signals from other nodes, and 3) the
thermal noise. Since the interfering signals come from other
nodes, we assume that the total interfering signal can be
treated as an additive noise process independent of the
thermal noise process. The received signal i during each bit
period can be expressed as
i :
sig
X
`2
,1
:
,
n
thermal
. 12
where :
sig
is the signal from the transmitter, :
,
is the signal
from an interfering node ,, and n
thermal
is the thermal
noise signal. In the following, we will derive these three
components.
Consider a link between the transmitter and the receiver
shown in Fig. 3. Assuming that a signal is attenuated with a
distance raised to the power , where is the pathloss
exponent, the power of the intended signal from the
transmitter as observed at the receiver can be written as [20]
1
r

c1
t
i
link
. 13
where
c
4
G
t
G
r
c
2
4
2
)
2
c
14
and 1
t
is the transmit power, G
t
and G
r
are the transmitter
and receiver antenna gains, )
c
is the carrier frequency, and c
is the speed of light. In this paper, we assume that the
antennas at the nodes are omnidirectional (G
t
G
r
1),
and the carrier frequency is in the unlicensed 2.4 GHz band.
Assuming a binary phase shift keying (BPSK) modulation,
there can be two cases for the amplitude of the received
signal: 1) :
sig

1
r
,1
b
p

4

1
b
p
if a +1 is transmitted and
2) :
sig

1
r
,1
b
p

4

1
b
p
if a -1 is transmittednote
that 1
b
is the bit energy of the received signal.
The thermal noise power can be written as follows:
1
thermal
1/T
0
1. 15
where 1 is the noise figure, / 1.38 10
23
J/K is
the Boltzmanns constant, T
0
is the room temperature
(T
0
300 K), and 1 is the transmission bandwidth. The
received thermal noise signal is simply n
thermal

1/T
0
1
p
.
Due to spatial invariance (torus assumption), without
any loss of generality, one could compute the amount of
interference experienced by a receiving node as if it were at
the center of the network. Fig. 3 illustrates a scenario where
the receiver is at the center of the network and the other
nodes are grouped in concentric square tiers. Consider a
potential interfering node , at a distance i
,
i
link
from the
receiver, where i
,
is a multiplicative factor which depends
on the position of node ,. For example, a node at the corner
of the first tier would have a multiplicative factor i
2
p
because it is at the distance
2
p
i
link
from the receiver. The
interference power from node , can be written as
1
int,

c1
t
i
,
i
link

1
r
i
,
. 16
For each interfering node ,, the amplitude of the
interfering signal can be classified into one of these three
cases: 1) :
,

1
int,
,1
b
p

1
b
,i
,
q
if a +1 is transmitted,
2) :
,

1
int,
,1
b
p

1
b
,i
,
q
if a -1 is transmitted,
and 3) :
,
0 if node , does not transmit. The probability
that an interfering node will transmit and cause interference
depends on the MAC protocol employed. Considering the
RESGO MAC protocol and assuming that each node
transmits packets with fixed length 1 (dimension: [b/pck]),
it can be shown that the interference probability is equal to
the probability that an interfering node transmits during a
vulnerable interval of duration 1,1
b
[17]. This probability
can be written as
3
j
tran
1 c
`
t
1
1
b
. 17
The link BER can now be computed. Since we are
considering a linear binary modulation, without loss of
generality, we can assume that the transmitter transmits
+1. Thus, in this case, :
sig

1
b
p
. Let us define a random
vector

o
int
f:
1
. :
2
. . . . . :
`2
g, where :
,
is the amplitude of
the signal of an interfering node , received at the receiver.
Note that

o
int
is random because each :
,
can take one of
these three different values with the probability given
below:
3. Actually, this is the probability that a node will transmit given that it
always has a packet to send. This is a pessimistic assumption, as there could
be occasions when a node has nothing to send. Thus, the actual probability
that a node will transmit should be lower. After a careful analysis, we have
found that the actual probability that a relay node will transmit does not
have a simple closed-form expression. This is partly due to the fact that the
arrival process of the relay node is no longer a Poisson process. Thus, we
believe that it is better to use a pessimistic assumption which allows us to
derive an expression for a transmission probability that provides insight
into the problem. Since we are using a pessimistic assumption, the actual
performance with the RESGO MAC protocol is expected to be better than
what is predicted by our analysis.
:
,

1
b
,i
,
q
with probability
1
2
j
tran
1
b
,i
,
q
with probability
1
2
j
tran
0 with probability1 j
tran
.
8
>
>
<
>
>
:
18
Assuming that the threshold for bit detection is placed at
0, the bit error probability can be written as
1fbit errorg BER
link

X
o
int
1
1
b
p

X
`2
,1
:
,
n
thermal
< 0j
o
int
( )
1

o
int
n o
.
19
where the summation is carried out over all possible
interference configurations given by

o
int
. Given the
random vector

o
int
, the conditional error probability can
be expressed as
1
1
b
p

X
`2
,1
:
,
n
thermal
< 0j
o
int
( )
Q
1
b
p

P
`2
,1
:
,
o
!
.
20
where o
1/T
0
,2
p
. Assuming that a bit detected erro-
neously at the end of a link is not corrected in successive
links, the BER at the end of a route with i
grid
links, denoted
as BER
grid
route
, can be written as
BER
grid
route
1 1 BER
link
i
grid
. 21
Since we assume that each intermediate node decodes the
signal before forwarding to the next node in a route, this is
regarded as decode-and-forward transmission. The case
where each intermediate node simply combines the
received signal and relay the signal without decoding can
also be considered following the approach in [21], [22].
Note that the size of the vector

o
int
increases as the
number of nodes in the network increases. Moreover, the
values that

o
int
can take (and, hence, its pmf) are of the order
of 3
`
. Thus, considering all nodes in the network as
potential sources of interference may not be practical.
However, our study shows that the dominant interference
signals come from the nodes in the first tier around the
receiver, and interference from nodes in other tiers are not
significant. In Fig. 4, we compare the route BER in two
scenarios: 1) interference signals are from nodes in the first
tier only and 2) interference signals are from nodes in both
the first tier and the second tier. The results are obtained
assuming that the number of nodes is ` 289, the transmit
power is 1
t
1 mw, the packet size is 1 1.000 b/pck, and
the packet transmission rate is `
t
0.5 pck/s. The con-
sidered data rates are 1
b
2 Mb/s and 1
b
100 kb/s. It
can be observed that the route BER in scenarios 1) and 2) are
not significantly different for both data-rate values. As a
result, without significant loss of accuracy, one can estimate
the route BER by considering only the interference from
nodes in the first tier.
To verify our analysis, we also compute the route BER
via Monte-Carlo simulations. In the simulation, each node
transmits packets according to the RESGO MAC protocol.
For each bit received by the receiver, the detection decision
is made based on the total received signal (i.e., the signal
from the transmitter, the interference signals, and the
thermal noise). If the received bit is different from the
original bit sent by the transmitter, it is declared as a bit
error. Through a Monte-Carlo approach, the link BER can
then be obtained by calculating the ratio between the
number of bits received in error and the total number of bits
received. Finally, the route BER can be obtained from the
link BER using (21). In Fig. 5, we compare the route BER
obtained from analysis with that obtained by simulation.
The parameters used are the same as those in Fig. 4. It can
be observed that both analytical and simulation results are
in good agreement.
4.1.2 Approximate Analysis
The link BER and the route BER can be rigorously
evaluated with the approach presented in Section 4.1.1.
However, in this section, we provide an approximation
approach which allows us to calculate the link BER and the
route BER more conveniently. In this approach, the link
BER is evaluated by assuming that, in each link, the total
interference signal can be modeled as a Gaussian process
Fig. 4. Route BER as a function of node spatial density, comparing the
case where only the interference from the first tier is considered and the
case where the interference from the first two tiers is considered.
Fig. 5. Route BER as a function of node spatial density. Both analytical
results and simulation results are compared.
with variance corresponding to the total average interference
power. While we used this approach in our earlier works
[10], [23], a more accurate analysis reveals that this
approach has limited validity. We now show how to
correctly use this approach. In order to do this, we first
evaluate the average interference power as observed by the
receiver. Since each node transmits with a common power
1
t
, if all nodes simultaneously transmit, it can be shown
that the total interference power experienced by the
receiver at the center of the network is [17]
1
grid
total
c1
t
1
grid
. 22
where
1
grid

4
1
i
link
X
imax
i1
4
i
2
p
i
X
i1
,1
8
i
2
,
2
p

1
" #
. 23
The first term in the summation corresponds to the four
nodes in each tier that are aligned with the receiver at the
center of the network, the second term in the summation
corresponds to contribution from the four nodes at the
corners of each tier at distance
2
p
ii
link
from the receiver,
and, finally, the last term in the summation corresponds to
the contribution of all the other nodes in each tier. Note that,
in each tier, there are eight nodes at distance
i
2
,
2
p
i
link
from the receiver, for , < i.
With the RESGO MAC protocol, each node transmits
with probability j
tran
and, thus, the average interference
power can be written as
E1
grid
int
j
tran
1
grid
total
1 c
`
t
1
1
b

1
grid
total
. 24
Considering BPSK signaling and assuming that the
interfering noise is Gaussian, the BER on each link of a
route can be written as [24]
BER
link
grid
Q
2c1
t
,i
link
1
thermal
E1
grid
int

s !
. 25
The route BER can then be evaluated using (21).
The route BER from the exact interference analysis
presented in Section 4.1.1 shows the existence of a BER
floor for increasing values of the node spatial density or
transmit power. A typical BER floor can be observed in
Fig. 5. This is intuitively expected because, after the network
becomes dense enough, increasing the node spatial density
(i.e., reducing the hop length) no longer improves the link
SNR, as the interfering nodes also become close enough to
the receiver. However, the route BER predicted by the
Gaussian assumption for the interference signal presents a
much lower BER floor. Via a careful analysis, the route BER
floor can be approximated with the following expression:
BER
floor
%
3i
grid
`
t
1
41
b
. 26
Motivated by the above considerations, one finds out
that an accurate approximate expression for the route BER
can be written as
BER
route

max 1 1 Q
2c1
t
,i
link
1
thermal
E1
grid
int

s ! " #
i
grid
0
@
1
A
.
3i
grid
`
t
1
41
b
8
<
:
9
=
;
.
27
In Fig. 6, to verify the accuracy of this approximation, we
compare the route BER obtained from the exact interference
analysis (e.g., using the approach in Section 4.1.1) with that
obtained from the approximate analysis (e.g., using (27)) for
two different values of data rate. The parameters used are
the same as those in Fig. 4. It can be observed that the route
BER obtained from the approximation closely matches the
route BER obtained from the exact interference analysis.
4.2 Random Topology
In the case of random topology, it is difficult to obtain exact
analytical expressions for the link BER and the route BER.
Therefore, we resort to a simulation approach for the
evaluation of the BER. For each communication link, a
transmitter and receiver pair is generated. The distance
between the transmitter and the receiver (i.e., \) is
randomly generated from the hop length distribution given
in (4). Then, a number of interfering nodes within the circle
of radius 2\ centered at the receiver are generated
according to a two-dimensional Poisson distribution. Their
positions are uniformly distributed. For each node, we
schedule transmissions of packets according to the RESGO
MAC protocol. For each bit received at the receiver, a
threshold-based decision is made. The detected bit is
declared as an error if it differs from the original bit sent
by the transmitter. The link BER is obtained by computing
the ratio of the number of erroneous bits and the total
number of received bits. To obtain the route BER, we
generate a route with i
rand
links and compute the route BER
as follows:
BER
rand
route
1
Y
i
rand
i1
BER
linki
. 28
where BER
linki
is the BER of the ith link of a route. The
average route BER is then obtained by computing the
Fig. 6. A comparison between the route BER obtained with the exact
interference analysis and the approximation.
arithmetic average of a large number of realizations of
BER
rand
route
.
5 OPTIMAL COMMON TRANSMIT POWER
In this section, we derive the optimal common transmit
power for ad hoc wireless networks with grid and random
topologies, respectively.
5.1 Optimal Common Transmit Power for Networks
with Square Grid Topology
With the exact BER analyzed in Section 4.1.1, one can
always numerically find the minimum transmit power
which would satisfy the desired BER threshold, provided that
the BER threshold is above the BER floor. However, it is
more insightful to derive an analytical expression for the
optimal transmit power. In this section, we will derive an
expression for the optimal transmit power for a given
maximum tolerable route BER from the approximate route
BER expression given in (27).
In a network with grid topology, the common transmit
power used by each node should be large enough so that
the BER at the end of a multihop route with an average
number of hops i
grid
, given by (5), is lower than the
maximum tolerable value, denoted as BER
th
. If the required
BER
th
is higher than BER
floor
, there exists a power which can
satisfy this requirement. This transmit power is such that
the following inequality must be satisfied:
1 1 BER
link
grid
i
grid
BER
th
. 29
Substituting expression (25) for BER
linkgrid
into (29), one can
rewrite (29) as
1 1 Q
2c1
t
,i
link
1
thermal
E1
grid
int

s ! " #
i
grid
BER
th
. 30
Substituting 1/T
0
1
b
for 1
thermal
, the expression given in (24)
for E1
grid
int
, and rewriting (30) in terms of 1
t
, one obtains the
following expression for the optimal transmit power:
1
t
1/T
0
1
b
2c
i
link
c 1 c
`
t
1
1
b

1
grid
!
1
dimension : Watt.
31
where

4
Q
1
1 1 BER
th

1,igrid
h i n o
2
32
and 1
grid
is given by (23). The expression given in (31)
corresponds to the optimal transmit power when
BER
floor
BER
th
for a given data rate 1
b
, node spatial density (i
link
1,

,
s
p
),
number of nodes in the network (i
grid

`
p
,2), antenna
gains, and carrier frequency (c depends on them).
5.2 Optimal Common Transmit Power for Networks
with Random Topology
In a network with two-dimensional Poisson node distribu-
tion, the hop length is random. Consequently, it is difficult
to find a closed-form expression for the optimal transmit
power. In this case, we resort to simulations. For a given
data rate and transmit power, we compute the average
route BER with the approach described in Section 4.2. The
same process is repeated for different combinations of data
rate and transmit power. At the end, we thus obtain the
average route BERs for different combinations of data rate
and transmit power, and these are used to obtain contours
of power-data rate pairs in correspondence to which the
route BER is equal to maximum tolerable BER
th
values.
6 PERFORMANCE METRICS
6.1 Node and Network Lifetime
Network lifetime is an important performance indicator for
wireless ad hoc and sensor networks. There are several
definitions of network lifetime. While network lifetime is
defined as the time to the first node failure in [25] and [26],
it is defined in terms of the fraction of surviving nodes in
the network in [27]. In this paper, following [25] and [26],
we consider the time to the first node failure as the network
lifetime (i.e., a worst-case approach).
In this section, we present a simple analysis for
computing the average lifetime of a node. We assume that
the RESGO MAC protocol is used and every node has an
initial finite battery energy denoted by 1
batt
. For a given
data rate 1
b
, the time taken to transmit one packet is 1,1
b
.
Therefore, the total amount of energy consumed per
transmitted packet can be written as
1
packet
1
t
1
1
b
dimension : Joules. 33
Since packets are transmitted with average rate `
t
, the
average energy depleted per second is simply `
t
1
packet
.
Finally, the total time it takes to completely exhaust the
initial battery energy can be written as
t
1
batt
`
t
1
packet
1
batt
1
b
`
t
11
t
dimension : sec. 34
Note that, due to uniform traffic assumption (i.e., all nodes
generate approximately the same amount of traffic load), on
average, all nodes exhaust their battery at the same time. In
addition, this simple analysis does not take into account the
energy consumed when a node is receiving and processing
packets. In reality, the lifetime of a node will be shorter than
what is predicted by our analysis. Recent studies have
shown that the energy consumption ratio when a device is
in idle mode, receiving mode, and transmitting mode is
1:1.2:1.68 (see, e.g., [28]). Therefore, transmitting is the most
expensive (from a battery consumption perspective)
activity. Power consumption in idle and receiving modes
can be taken into account by properly extending the
proposed approach.
6.2 Effective Transport Capacity
The effective transport capacity (dimension: [b-m/s]) is an
important wireless network performance metric which
simultaneously captures the amount of information gener-
ated and the distance over which it can be transported in
the network [23]. As is evident from (21), the BER at the end
of a multihop route increases as the distance traversed by
an information bit increases, i.e., as the number of hops
increases. Consequently, the distance that an information
bit can traverse while maintaining sufficiently low BER may
be less than the length of a multihop route with an average
number of hops (i.e., a message could reach the destination,
but the final BER could be so high that the destination could
not decode the received message correctly). The following
QoS condition, in terms of tolerable BER at the end of a
multihop route with i hops, formalizes the requirement of
sufficiently low route BER:
BER
i
route
BER
th
. 35
where BER
i
route
is the BER at the end of a i-hop route. We
refer to the distance which a bit can traverse, while
satisfying a desired BER QoS condition, as maximum
sustainable distance. The effective transport capacity of a
single route can then be defined as
C
sr
Te

4
`
t
1 mind
s
. d
a
. 36
where d
s
is the maximum sustainable distance and d
a
is the
distance associated with an average multihop route. Note
that, in (36), the traffic transmission rate is used, whereas
one should consider a traffic generation rate which
represents the average rate of information flowing in a
route. However, since we have assumed in this paper that a
node always has packets to send, the average flow rate in a
route, in this case, is equal to the average transmission rate
`
t
. The value mind
s
. d
a
can be interpreted as the average
sustainable distance. It represents the distance that a bit, on
average, can traverse. A fundamental underlying assump-
tion in (36) is that only the source node of a route contributes
effective information (on average, a source node trans-
mits `
t
1 bits every second). In this sense, the intermediate
nodes act as relay nodes, but they do not contribute to the
effective transport capacity in terms of information bits.
However, relay nodes do contribute to the effective
transport capacity in the sense of increasing the distance
traversed by each bit transmitted by the source.
In this paper, the effective transport capacity of the
network is obtained by adding the single-route effective
transport capacities of all the disjoint routes in the network.
4
Assuming that each route has an average number of hops,
the number of disjoint routes corresponding to a scenario
where each node belongs to a particular route (either as the
source, the destination, or a relay) in networks with grid
and random topologies can be estimated as `,i
grid
and
`,i
rand
, respectively. The extension of the proposed
analysis to a scenario where multihop routes can cross
can be done by following the approach in [29].
In the following section, we derive analytical expressions
for d
s
and d
a
in networks with square grid topology and
networks with two-dimensional Poisson topology, respec-
tively. The corresponding effective transport capacities in
both cases can be obtained using (36).
6.2.1 Square Grid Topology
First, we compute the maximum sustainable distance. In the
case of square grid topology, the route BER at the end of a
generic i-hop route can be obtained by replacing i
grid
in (27)
with i. Thus, (35) will be satisfied if
max 1 1 Q
2c1
t
,i
link
1
thermal
E1
grid
int

s ! " #
i
!
.
3i`
t
1
41
b
( )
BER
th
.
37
Solving (27) for i, the maximum sustainable number of
hops can be written as
i
grid
s
min
ln1 BER
th
ln1 BER
linkgrid
.
41
b
BER
th
3`
t
1
& '
. 38
Since all hops in a network with grid topology have the
same length i
link
, the maximum sustainable distance is
simply the product of the maximum sustainable number of
hops and the hop length:
d
grid
s
i
grid
s
i
link
. 39
Similarly, the average route distance in a grid topol-
ogy, i.e., d
grid
a
, can be obtained as the product of the
average number of hops i
grid
with the hop length i
link
,
i.e., d
grid
a
i
grid
i
link
.
6.2.2 Random Topology
As in the case of a network with grid topology, in order to
obtain the single-route (and the aggregate) effective trans-
port capacity of a network with random topology, we need
to compute the maximum sustainable distance and the
average route distance. In this case, however, the maximum
sustainable distance is obtained via computer simulations.
We construct a route by recursively generating links whose
lengths are obtained according to the CDF given in (4). After
the generation of a link, the route BER is then computed and
compared with the threshold value, i.e., BER
th
. If the BER is
lower than BER
th
, a newlink will be generated and added to
the route. This procedure is repeated until the route BER
exceeds BER
th
. The maximum sustainable distance is
obtained by adding the lengths of the links of the formed
multihop route. A sufficiently large number of multihop
routes are generated and the average sustainable distance is
computed by averaging over the maximum sustainable
distances corresponding to the route realizations.
What now remains to be computed is the average route
distance in a random topology, i.e., d
rand
a
. The average
number of hops in a route between a source and a
destination is i
rand
. To find the average distance, one can
compute the expected value of the route distance in a
i
rand
-hop route, that is,
d
rand
a
E
X
irand
i1
\
i
" #
. 40
where \
i
is the hop length of the ith link. Since i
rand
is a
deterministic quantity, (40) can be simplified to
d
rand
a
i
rand
E \ . 41
4. The disjoint routes assumption is particularly assumed here for
simplicity in calculating the networks effective transport capacity.
However, other results derived in this paper hold regardless of whether
routes are disjoint or not.
7 RESULTS AND DISCUSSION
Numerical results, along with their implications, are
presented and discussed in this section. The values of the
major network parameters are given in Table 1 unless
specified otherwise.
7.1 Optimal Transmit Power and Data Rate
In Fig. 7, the optimal common transmit power is shown as a
function of the data rate. The optimal power-data rate
curves are shown for various values of BER
th
. For each case
of BER
th
, both the optimal transmit power computed
numerically with the exact interference analysis and the
optimal transmit power computed with the approximation
given in (31) are shown. It can be observed that the optimal
transmit power increases as the data rate increases. This can
be explained as follows: Although transmitting packets at a
higher data rate reduces the vulnerable time (and, hence,
smaller interference), increasing the data rate (i.e., band-
width) also increases the thermal noise. Therefore, the
minimum transmit power required to sustain the network
connectivity has to increase accordingly. We also observe a
close match between the optimal transmit power predicted
with the approximation and the one obtained numerically
from the exact interference analysis.
In addition, it can be observed from Fig. 7 that there is a
critical data rate, beyond which the desired BER
th
cannot be
satisfied for any transmit power. The critical data rate occurs
at the point where the BER
floor
for that particular data rate
becomes higher than the desired BER
th
. Consequently, for
an uncoded transmission and without retransmission, no
transmit power can achieve the desired BER threshold.
The optimal transmit power is also minimized at the data
rate near the critical point. This suggests that the data rate
also plays an important role in the design of wireless ad hoc
and sensor networksthat is, for a given node spatial
density, if the data rate is carefully chosen, the transmit
power can be minimized, prolonging the networks lifetime.
A scenario with random topology is shown in Fig. 8 for
the case where BER
th
10
3
. Similar to the case of grid
topology, the optimal transmit power increases as the data
rate increases. However, in the case of random topology,
as expected, a much higher transmit power is required to
sustain connectivity. This is due to the fact that the hop
length is random. In other words, a long hop is likely to
be present in a multihop route, and this significantly
reduces the route BER [15]. In addition, in the case of
random topology, higher data rates are required to
support connectivity.
The optimal transmit power shown in Fig. 7 is computed
based on the assumption that there is no packet retransmis-
sion on each link of a route. In some applications, packet
retransmission may also be used. In Fig. 9, we compare the
optimal transmit power in three different scenarios: 1) no
retransmissions on each link, 2) one retransmission is
allowed on each link, and 3) five retransmissions are
allowed on each link. The BER
th
considered in this case is
10
3
. It can be observed that allowing retransmission can
decrease the minimum transmit power required to keep the
network connected. This is because retransmission im-
proves the BER and, thus, a lower transmit power is
required to get a packet to the destination with the desired
quality. It can also be observed that the critical data rate
shifts leftward as the number of retransmissions increase.
Thus, by allowing retransmission, the network connectivity
can be supported at lower data rates.
TABLE 1
Major Network Parameter Values Used
in Considered Networking Scenarios
Fig. 7. Optimal common transmit power in a network with square grid
topology. Only interference from nodes in the first tier around the
receiver is considered.
Fig. 8. Optimal common transmit power in a network with random
topology and BER
th
10
3
.
7.2 Optimal Transmit Power and Node Spatial
Density
Fig. 10 illustrates the optimal transmit power as a function of
the node spatial density. In this scenario, the data rate is
fixed at 1
b
4 Mb/s, and the considered BER
th
values are
10
2
and 10
3
, respectively. Both the optimal transmit power
computed numerically with the exact interference analysis
and the optimal transmit power computed with the
approximation given in (31) are compared. It can be
observed that the optimal transmit power decreases linearly
in the log scale as the node spatial density increases. This can
be explained as follows: When the network is sparse (i.e.,
low node spatial density), the minimum transmit power has
to be increased in order to preserve connectivity. However,
as the network becomes denser, the nodes are closer to their
neighbors and, therefore, the power required to sustain
connectivity decreases.
Note that the BER
floor
for the scenario considered in
Fig. 10 is fixed because the data rate 1
b
, the number of
nodes `, the traffic transmission rate `
t
, and the packet
length 1 are fixed. As long as the desired BER
th
is greater
than this BER
floor
, there is always a minimum power to
sustain the connectivity at any node spatial density. On the
other hand, in the case where BER
th
is less than the
BER
floor
, no transmit power can satisfy the desired con-
nectivity criteria at any node spatial density, unless forward
error correction or retransmission techniques are used.
7.3 Effects of Strong Propagation Pathloss
In a realistic communication environment, some links may
experience higher pathloss than others due to impairments
such as shadowing [20]. In order to study how pathloss
affects the optimal common transmit power, next, we
consider scenarios where some links in a multihop route
are characterized by higher pathloss than the others. We
study the impact of these high pathloss links on the optimal
transmit power by varying the number of these links in a
multihop route with an average number of hops.
Fig. 11 illustrates the optimal common transmit power as
a function of the data rate in a scenario where the RESGO
MAC protocol is used and the network topology is a square
grid. Four scenarios are considered:
1. all links in an average multihop route are character-
ized by a pathloss exponent 2,
2. the i 1 link is characterized by a pathloss
exponent 3, whereas the other links in the route
are still characterized by 2,
3. the i 3 links are characterized by a pathloss
exponent 3, whereas the other links in the route
are still characterized by 2, and
4. all links in the route are characterized by a pathloss
exponent 3.
In the scenarios where the pathloss exponents for different
links are not the same, attenuation of the signals originating
from interfering nodes will also vary. If the pathloss for
each interfering node, relative to the receiver, were known,
then the exact interference power could be computed.
However, this would only correspond to a specific realiza-
tion of the network, where one needs to determine the
propagation loss on each individual link from any node to
any other nodes.
In order to estimate the interference power, we assume,
for simplicity, that the interference power from interfering
topology and BER
th
10
3
. Different number of retransmissions are
compared.
Fig. 10. Optimal common transmit power in a network with grid topology
as a function of node spatial density.
topology for various configurations of the links pathloss exponents. In
this case, the node spatial density is ,
s
10
5
m
2
and BER
th
10
3
.
nodes attenuate with a common pathloss exponent
c
, given
by the following heuristic expression:
c

i
grid
i
i
grid
2
i
i
grid
3. 42
In other words, (42) for the common pathloss exponent
corresponds to a weighted average between 2 and 3. The
intuition behind (42) is that, if the number of links per route
with 2 is large, then the pathloss exponent of each
interfering node to the receiver should be close to 2; on the
other hand, if the number of links with 3 is large, then
the pathloss exponent of each interfering node to the
receiver should be close to 3. We emphasize that the
expression in (42) is only a heuristic. However, the results
obtained with this analysis provide, at least trendwise,
meaningful insights.
Generally, it can be observed from Fig. 11 that the power
required to sustain network connectivity increases as the
number of links with 3 increases. Moreover, the
transmit power has to be increased substantially even if
only one link in a multihop route experiences higher
pathloss than the others. This has a significant implication.
It suggests that the use of a common transmit power will
work well in situations where each link experiences
relatively the same level of propagation loss. If it is highly
likely that at least one link per route is going to experience
unusually high pathloss, then it is not efficient to use a
common transmit power because all nodes have to
significantly raise their transmit power to recover from
the degradation caused by a few high pathloss links. A
possible solution to this problem is to modify the routing
protocol in such a way that, at the moment of route
creation, link quality is selected as one of the routing
criteria [30], [31]. With this capability, a routing protocol
may be able to select a multihop route which bypasses high
pathloss links. Another potential solution is obviously
considering local power control [4], [5], [6]. These possible
extensions are currently under investigation.
7.4 Connectivity Robustness to Node Spatial
Density Changes
The optimal transmit power analyzed earlier is the mini-
mum transmit power, which is sufficient to maintain
network connectivity at a particular node spatial density.
In some situations, the node spatial density of a network can
change over time, for example, when nodes join/leave the
network or when the batteries of some nodes are exhausted.
A network can lose connectivity immediately if the node
spatial density is lower than the initial value for which the
power is optimally chosen. In order to provide robustness
against a possible node spatial density reduction, nodes may
transmit at a power level higher than the required minimum
level. In this section, we evaluate the connectivity robustness
as the network becomes sparser. We define the initial density
at which the network starts forming as ,
s
0

4
`
0
,, where `
0
is the initial number of nodes in the network. Keeping the
network area constant, the node spatial density decreases if
the number of nodes becomes lower than `
0
.
5
We denote the
decrease in the number of nodes as `
4
`
0
`
f
, where `
f
is the final number of nodes after some nodes disappear from
the networksome nodes could either move out of the
network or their batteries could be exhausted.
Fig. 12a shows a plot of ` as a function of the
transmit power. In this scenario, the initial number of
nodes is `
0
289 and the network surface area is
10
8
m
2
. The minimum transmit power required to
sustain the network connectivity for this initial node
spatial density is 1
t
0
1.6 mW. In this figure, for each
transmit power, the node spatial density and the number
of nodes `
f
at which the network can still maintain
connectivity is calculated.
This allows one to obtain `, which, in this figure, can
be interpreted as the connectivity robustness against a
change in terms of node spatial density. For example, `
100 indicates that a network can still maintain connectivity
even if 100 nodes disappear.
It can be observed from Fig. 12a that, as the transmit
power increases, ` increases.
6
This is intuitively expected
because, for higher transmit power, the network can
tolerate lower node spatial densities, i.e., it can work even
if the network becomes sparser. However, as the transmit
power level increases, the rate of gain, in terms of
connectivity robustness, asymptotically decreases.
While connectivity robustness can be increased with
higher transmit power, this comes at the expense of the
nodes and networks lifetimes. Fig. 12b illustrates the
behavior of the lifetime of a node as a function of the
transmit power. In particular, the lifetime of a node is given
by (34). As expected, the node lifetime decreases as the
transmit power increases to support communication in a
sparser network.
Fig. 12. (a) Increase in connectivity level and (b) lifetime of a node as
functions of transmit power in a network with grid topology.
5. The proposed analysis can be straightforwardly extended to a scenario
where the number of nodes is fixed and the network area increases.
6. The decrease in the number of nodes (`) is computed as if the nodes
are still in a square grid topology after ` nodes disappear from the
network. In reality, nodes will probably disappear from the network in a
random fashion, and the topology is likely to be random after some nodes
disappear. Thus, the power required to maintain network connectivity is
expected to be much higher than what is shown in Fig. 12. However, the
results shown in Fig. 12 provide a meaningful insight in terms of the trend
involved.
As shown in Fig. 12a and Fig. 12b, it is clear that there is a
trade-off between connectivity increase and nodes long-
evity. This trade-off is clearly shown in Fig. 13, where
network lifetime is shown as a function of the connectivity
robustness. It can be observed that the slope of the curve
becomes steeper as ` increases. This suggests that trading
network lifetime for connectivity is beneficial when the
level of connectivity is small; however, the benefit decreases
at high values of `. In other words, to have high
protection against node spatial density changes, significant
network lifetime has to be sacrificed.
The increased robustness against node spatial density
reduction is also exhibited by the behavior of the effective
transport capacity. Fig. 14 shows the behavior of the
effective transport capacity, in a scenario with square grid
topology, as a function of the number of nodes. The results
are obtained assuming that the data rate is 1
b
4 Mb/s,
the network area is 10
8
m
2
, and the desired level of
BER is BER
th
10
3
. By keeping the area constant, an
increase of the number of nodes corresponds to an increase
of the node spatial density. Three levels of transmit powers
are considered: 1) 1
t
0
1.6 mW, 2) 2 1
t
0
, and 3) 5 1
t
0
.
Considering the effective transport capacity curve, one is
mostly interested in the region where the network is
connected. This is the region where an information bit can
be transported over an average route while satisfying the
QoS constraint on the maximum tolerable BER, i.e., BER
th
[23]. The solid curve in Fig. 14 indicates the effective
transport capacity in the region where the network has
connectivity. The connectivity region for each considered
transmit power starts from the point where each dashed
curve reaches the solid curve and ends at the knee of the
curve (e.g., around ` 425). It can be seen from Fig. 14
that, when the transmit power is increased from 1
t0
to
2 1
t
0
, the connectivity region of the network increases.
More specifically, the network gains connectivity at `
289 in the case where the transmit power is 1
t
0
, whereas it
gains the connectivity at a much lower number of nodes
(e.g., around ` 160) in the case where the transmit power
is 2 1
t
0
. Hence, transmitting at a power higher than the
minimum required level can increase the connectivity
range, in the sense that communications can be supported
also in sparser networks. In other words:
. if the network is already connected, increasing the
transmit power does not increase the effective
transport capacity;
. if the network is not connected, increasing the
transmit power might make it connected, increasing
dramatically the effective transport capacity.
7.5 Practical Determination of the Optimal Transmit
Power
In this paper, we have presented an analytical approach to
determine the optimal common transmit power for network
connectivity. Next, we briefly discuss how this optimal
transmit power can be determined practically. There are
at least two possible strategies based on the capability of the
devices. First, in a wireless network where a device can
operate with only a single power level, the optimal transmit
power can be determined at the design phase. For example,
one can estimate the number of nodes and the size of the
coverage area that a wireless sensor network will be
deployed across and then configure a device such that it
operates at the optimal transmit power. Alternatively, if the
transmit power of a device is preconfigured at the
manufacturing stage, in the network planning phase, one
can determine the minimum node spatial density required
for this transmit power. This enables a network designer to
determine the minimum number of nodes to be deployed in
the coverage area.
Second, if each device has an option of changing its
transmit power, a distributed power control algorithm can
be used. By adapting the COMPOW algorithm [1], the
minimum common transmit power for which the route BER
satisfies the desired QoS can be determined. Assuming that
each device has several discrete power levels, this can be
done as follows: Starting from the maximum power level,
each node creates a routing table corresponding to this
power level. This allows each node to identify all the
possible destinations that it could reach. For each destina-
tion, a node sends a probing packet to check the route BER
at the destination. The same process is repeated for the
other available power levels. The minimum power which
yields the same routing table as if the maximum power
were used and satisfies the required route BER is the
optimal power level.
Fig. 13. Lifetime of a node as a function of connectivity.
Fig. 14. Effective transport capacity in a network with grid topology.
A careful reader may note that our approach does not
take an opportunistic longer hop routing approach,
although it may become available when the transmit power
becomes larger. However, we emphasize that, although the
transmit power is enough to reach a node further than the
nearest neighbor, it may still not be enough to reach the
destination with the desired level of BER. Certainly, there
are trade-offs between choosing the nearest neighbor and
choosing the furthest node in a transmission range. The
latter is similar to selecting a route with the lowest number
of hops to the destination. For more detail on the trade-offs
between these two routing strategies, we refer the interested
reader to [30], [31].
8 RELATED WORK
There are a number of distributed power control algorithms
reported in the literature, and it is important to emphasize
that they are designed to achieve different objectives. In
[32], Foschini and Miljanic proposed a distributed power
control algorithm for a cellular network such that the
minimum signal-to-interference ratio requirement per user
is met. An algorithm proposed by Bambos and Kandukuri
[33] attempted to balance the power and delay trade-off by
using local information such as last transmitted power
level, last interference level, and backlog at the transmitter
to determine the appropriate power to transmit in the next
time slot. In [34], Monk et al. proposed an algorithm which
aims at improving channel efficiency. In [35], Elbatt et al.
proposed a power management scheme for improving the
end-to-end network throughput. In [25], Chang and
Tassiulas proposed a routing algorithm which aims at
balancing the energy consumption among the nodes.
However, none of these distributed power control algo-
rithms were designed with the objective of ensuring
network connectivity.
For our purpose, a distributed power control algorithm
should adjust the power such that the minimum power is
used to achieve network connectivity. Distributed power
control algorithms that are designed with network con-
nectivity and network longevity as their objectives also exist.
Some examples of these algorithms are [3], [6], and [27]. One
example of distributed power control algorithms, called
COMPOW, for determining the minimum common transmit
power that keeps the network connected is proposed in [1].
In this algorithm, it is assumed that each device has several
discrete power levels to use. The COMPOW algorithm
works as follows: Starting from the maximum transmit
power level, each node creates a routing table corresponding
to this power level. Thus, each node can identify all possible
destinations in the network that it could reach. The same
process is repeated for every available power level. The
common power level to use is the lowest power level that
enables each node to reach every destination as if the
maximum power were used. By adapting the COMPOW
algorithm, the minimum common transmit power for which
the route BER satisfies the desired QoS can be determined.
However, to the best of our knowledge, none of the previous
studies has considered the network connectivity from a
communication-theoretic viewpoint [16], [36], where the
desired QoS in terms of BER at the end of a multihop route is
enforced, as presented in this paper.
A similar analysis of BER at the end of a multihop route,
as presented in this paper, also exists in the literature. In
[21], Boyer et al. provide a detailed analysis of the route
BER in both the decoded relaying multihop channel case
and the amplified relaying multihop channel case. In our
analysis, we only consider the decoded relaying channel
casethat is, we assume that each intermediate node
decodes a received bit before forwarding to the next node
in the route. However, our BER analysis also includes
multiple access interference, whereas this is not considered
in [21].
9 CONCLUSIONS
In this paper, we have investigated the optimal common
transmit power for wireless sensor networks. In particular,
the optimal common transmit power has been defined as
the minimum transmit power sufficient to preserve network
connectivity. As opposed to following the conventional
graph-theoretic approach, we consider connectivity from a
more realistic viewpoint by taking into account the
characteristics of a wireless communication channel. Our
study shows that, for a given data rate and a given
maximum tolerable BER at the end of a multihop route,
there exists an optimal transmit power. Moreover, for a
given value of the maximum tolerable route BER, there
exists a global optimal data rate for which the optimal
common transmit power is the minimum possible. This
suggests that the data rate, if chosen carefully, can
guarantee significant savings in terms of transmit power,
prolonging the battery life of devices and network lifetime.
This is useful in designing wireless sensor networks. Since
the node spatial density of a sensor network is typically
known a priori (i.e., the number of sensors and the coverage
area are known), a network designer can determine the
optimal power and the optimal data rate to operate.
Conversely, in the case where the transmit power and the
data rate of a sensor node is preconfigured, a network
designer can use the approach presented in this paper to
determine the minimum node spatial density required to
keep the network connected.
However, caution should be exercised when using the
common power control scheme proposed in this paper, as
its performance is significantly affected by the variation in
propagation pathloss. In other words, the common power
control scheme performs well in the case where each link in
the network experiences similar propagation pathloss.
When variation in propagation pathloss occurs, the transmit
power needs to be increased substantially even though
there is only one link per route which experiences high
pathloss. A possible solution to this problem is to add
intelligence into the routing protocol. More precisely, a
routing protocol that takes link quality as one of its route
selection criteria could be an attractive option. However, if
it is not possible to avoid a high propagation loss link by
means of routing, local power control may be required.
In addition, there are trade-offs between robustness to
change in network connectivity and network longevity.
Considering a transmit power higher than the optimal
value helps preserve connectivity of networks in the
presence of node spatial density change. However, this
comes at the expense of node/network lifetime. Moreover,
in a scenario where the network is connected, our results
show that increasing the transmit power does not increase
the effective transport capacity at all.
Finally, the analytical framework presented in this paper
can be further extended to study other interesting scenarios.
For example, one may extend this work to estimate the
optimal transmit power for other MAC and routing
protocols. The channel model can also be modified to take
into account multipath fading.
ACKNOWLEDGMENTS
The authors are grateful to the three anonymous reviewers
of this paper and the associate editor, Professor P.R. Kumar
of the University of Illinois at Urbana Champaign, whose
constructive comments and suggestions helped to improve
the manuscript substantially. This work was supported in
part by a seed grant from Cylab of Carnegie Mellon
University.
REFERENCES
[1] S. Narayanaswamy, V. Kawadia, R.S. Sreenivas, and P.R. Kumar,
Power Control in Ad-Hoc Networks: Theory, Architecture,
Algorithm and Implementation of the COMPOW Protocol, Proc.
European Wireless 2002 Next Generation Wireless Networks: Technol-
ogies, Protocols, Services, and Applications, pp. 156-162, Feb. 2002.
[2] S. Agarwal, R. Katz, S.V. Krishnamurthy, and S.K. Dao,
Distributed Power Control in Ad-Hoc Wireless Networks, Proc.
IEEE Intl Symp. Personal, Indoor, and Mobile Radio Comm. (PIMRC),
vol. 2, pp. F59-F66, Sept. 2001.
[3] R. Ramanathan and R. Rosales-Hain, Topology Control of
Multihop Wireless Networks Using Transmit Power Adjustment,
Proc. IEEE Conf. Computer Comm. (INFOCOM), vol. 2, pp. 404-413,
Mar. 2000.
[4] T. Elbatt and A. Ephremides, Joint Scheduling and Power
Control for Wireless Ad Hoc Networks, IEEE Trans. Wireless
Comm., vol. 3, no. 1, pp. 74-85, Jan. 2004.
[5] R.L. Cruz and A.V. Santhanam, Optimal Routing, Link Schedul-
ing and Power Control in Multi-Hop Wireless Networks, Proc.
IEEE Conf. Computer Comm. (INFOCOM), vol. 1, pp. 702-711, Apr.
2003.
[6] V. Rodoplu and T. Meng, Minimum Energy Mobile Wireless
Networks, IEEE J. Selected Areas in Comm., vol. 17, no. 8, pp. 1333-
1344, Aug. 1999.
[7] C. Bettstetter and J. Zangl, How to Achieve a Connected Ad Hoc
Network with Homogeneous Range Assignment: An Analytical
Study with Consideration of Border Effects, Proc. IEEE Intl
Workshop Mobile and Wireless Comm. Network, pp. 125-129, Sept.
2002.
[8] C.-C. Tseng and K.-C. Chen, Power Efficient Topology Control in
Wireless Ad Hoc Networks, Proc. IEEE Wireless Comm. and
Networking Conf. (WCNC), vol. 1, pp. 610-615, Mar. 2004.
[9] Q. Dai and J. Wu, Computation of Minimal Uniform Transmis-
sion Power in Ad Hoc Wireless Networks, Proc. IEEE Intl Conf.
Distributed Computing Systems Workshops (ICDCS), pp. 680-684,
May 2003.
[10] S. Panichpapiboon, G. Ferrari, and O.K. Tonguz, Optimal
Common Transmit Power in Ad Hoc Wireless Networks, Proc.
IEEE Intl Performance, Computing, and Comm. Conf. (IPCCC),
pp. 593-597, Apr. 2005.
[11] I. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, A
Survey on Sensor Networks, IEEE Comm. Magazine, vol. 40, no. 8,
pp. 102-114, Aug. 2002.
[12] B. Schrick and M.J. Riezenman, Wireless Broadband in a Box,
IEEE Spectrum, vol. 39, no. 6, pp. 38-43, June 2002.
[13] G. Ferrari, S. Panichpapiboon, N. Wisitpongphan, R. Chokshi, and
O.K. Tonguz, Impact of Mobility on the BER Performance of
Multi-Hop Ad Hoc Wireless Networks, Proc. IEEE Global
Telecomm. Conf. (GLOBECOM), pp. 882-886, Nov. 2004.
[14] C.E. Perkins, Ad Hoc Networking. Addison-Wesley, 2001.
[15] S. Panichpapiboon, G. Ferrari, and O.K. Tonguz, Sensor Net-
works with Random versus Uniform Topology: MAC and
Interference Considerations, Proc. IEEE Vehicular Technology Conf.
(VTC), pp. 2111-2115, May 2004.
[16] O.K. Tonguz and G. Ferrari, Ad Hoc Wireless Networks: A
Communication-Theoretic Perspective. Wiley, 2006.
[17] G. Ferrari and O.K. Tonguz, Performance of Ad Hoc Wireless
Networks with Aloha and PR-CSMA MAC Protocols, Proc. IEEE
Global Telecomm. Conf. (GLOBECOM), pp. 2824-2829, Dec. 2003.
[18] N. Abramson, The Throughput of Packet Broadcasting Chan-
nels, IEEE Trans. Comm., vol. COM-25, no. 1, pp. 117-128, Jan.
1977.
[19] O.K. Tonguz and G. Ferrari, Is the Number of Neighbors in
Ad Hoc Wireless Networks a Good Indicator of Connectivity?
Proc. Intl Zurich Seminar Comm. (IZS 04), pp. 40-43, Feb. 2004.
[20] T.S. Rappaport, Wireless Communications Principles and Practice.
Upper Saddle River, N.J.: Prentice-Hall, 1996.
[21] J. Boyer, D. Falconer, and H. Yanikomeroglu, Multihop Diversity
in Wireless Relaying Channels, IEEE Trans. Comm., vol. 52,
no. 10, pp. 1820-1830, Oct. 2004.
[22] J. Boyer, D.D. Falconer, and H. Yanikomeroglu, On the
Aggregate SNR of Amplified Relaying Channel, Proc. IEEE
Global Telecomm. Conf. (GLOBECOM), vol. 5, pp. 3394-3398, Nov.
2004.
[23] G. Ferrari and O.K. Tonguz, MAC Protocols and Transport
Capacity in Ad Hoc Wireless Networks: Aloha versus PR-CSMA,
Proc. IEEE Military Comm. Conf. (MILCOM), pp. 1311-1318, Oct.
2003.
[24] S. Haykin, Communication Systems, fourth ed. John Wiley & Sons,
2001.
[25] J. Chang and L. Tassiulas, Energy Conserving Routing in
Wireless Ad-Hoc Networks, Proc. IEEE Conf. Computer Comm.
(INFOCOM), vol. 1, pp. 22-31, Mar. 2000.
[26] I. Kang and R. Poovendran, Maximizing Static Network Lifetime
of Wireless Broadcast Ad Hoc Networks, Proc. IEEE Intl Conf.
Comm. (ICC), vol. 3, pp. 2256-2261, May 2003.
[27] R. Wattenhofer, L. Li, P. Bahl, and Y. Wang, Distributed
Topology Control for Power Efficient Operation in Multihop
Wireless Ad Hoc Networks, Proc. IEEE Conf. Computer Comm.
(INFOCOM), vol. 3, pp. 1388-1397, Apr. 2001.
[28] B. Chen, K. Jamieson, H. Balakrishnan, and R. Morris, Span: An
Energy-Efficient Coordination Algorithm for Topology Mainte-
nance in Ad Hoc Wireless Networks, ACM Wireless Networks,
vol. 8, no. 5, pp. 481-494, 2002.
[29] S. Panichpapiboon, G. Ferrari, N. Wisitpongphan, and O.K.
Tonguz, Route Reservation in Ad Hoc Networks: Is It a Good
Idea? Proc. IEEE Wireless Comm. and Networking Conf. (WCNC),
pp. 2045-2050, Mar. 2005.
[30] N. Wisitpongphan, G. Ferrari, S. Panichpapiboon, J. Parikh, and
O.K. Tonguz, QoS Provisioning Using BER-Based Routing for
Ad Hoc Wireless Networks, Proc. IEEE Vehicular Technology Conf.
(VTC), May 2005.
[31] G. Ferrari, S.A. Malvassori, M. Bragalini, and O.K. Tonguz,
Physical Layer-Constrained Routing in Ad-Hoc Wireless Net-
works: A Modified AODV Protocol with Power Control, Proc.
Intl Workshop Wireless Ad-Hoc Networks (IWWAN 05), May 2005.
[32] G. Foschini and Z. Miljanic, A Simple Distributed Autonomous
Power Control Algorithm and Its Convergence, IEEE Trans.
Vehicular Technology, vol. 42, no. 4, pp. 641-646, Nov. 1993.
[33] N. Bambos and S. Kandukuri, Power-Controlled Multiple Access
Schemes for Next-Generation Wireless Packet Networks, IEEE
Wireless Comm. Magazine, vol. 9, no. 3, pp. 58-64, June 2002.
[34] J. Monk, V. Bharghavan, and W. Hwu, A Power Controlled
Multiple Access Protocol for Wireless Packet Networks, Proc.
IEEE Conf. Computer Comm. (INFOCOM), pp. 219-228, Apr. 2001.
[35] T. Elbatt, S. Krishnamurthy, D. Connors, and S. Dao, Power
Management for Throughput Enhancement in Wireless Ad-Hoc
Networks, Proc. IEEE Intl Conf. Comm. (ICC), vol. 3, pp. 1506-
1513, June 2000.
[36] O.K. Tonguz and G. Ferrari, A Communication-Theoretic
Approach to Ad Hoc Wireless Networking, Proc. 2006 Intl
Workshop Wireless Ad-Hoc Networks (IWWAN 06), Invited Paper,
June 2006.
Sooksan Panichpapiboon (S05) received the
BS and MS degrees in electrical and computer
engineering from Carnegie Mellon University
(CMU), Pittsburgh, Pennsylvania, in 2000 and
2002, respectively. He is currently a PhD
candidate in the Electrical and Computer En-
gineering (ECE) Department at CMU. His
research interests include ad hoc wireless net-
works, sensor networks, medium access control,
and performance modeling. He is a student
member of the IEEE.
Gianluigi Ferrari (S97-M03) received the
Laurea degree (5-year program) (summa cum
laude) and the PhD degree in electrical engi-
neering from the University of Parma, Italy, in
October 1998 and January 2002, respectively.
From July 2000 to December 2001, he was a
visiting scholar at the Communication Sciences
Institute, University of Southern California, Los
Angeles. Since 2002, he has been a research
professor in the Department of Information
Engineering, University of Parma. Between 2002 and 2004, he visited
the Electrical and Computer Engineering Department at Carnegie
Mellon University, Pittsburgh, Pennsylvania, several times as a research
associate. He has published more than 70 papers in leading interna-
tional conferences and journals. He is a coauthor of the books Detection
Algorithms for Wireless Communications, with Applications to Wired and
Storage Systems (John Wiley & Sons, 2004), Teoria della Probabilita e
Variabili Aleatorie con Applicazioni (McGraw-Hill, 2005), and Ad Hoc
Wireless Networks: A Communication-Theoretic Perspective (John
Wiley & Sons, 2006). He is listed in Marquis Whos Who in the World
and Whos Who in Science and Engineering, Madisons Whos Who in
the World, and in 2000 Outstanding Intellectuals of the 21st Century
(International Biographical Centre, UK, 2005). His research interests
include digital communication systems design, adaptive signal proces-
sing (with particular emphasis on iterative detection techniques for
channels with memory), information theory, and ad hoc wireless
networking. He was a corecipient of a Best Student Paper Award at
the 2006 International Workshop on Wireless Ad-Hoc Networks
(IWWAN 06). He acts as a frequent reviewer for many international
journals and conferences. He also acts as a technical program member
for several international conferences (IEEE ICC, IEEE WirelessCom,
IEEE IWCMC, Accessnets, IEEE WCNC, IWWAN, and IASTED CSA).
He is a member of the IEEE.
Ozan K. Tonguz received the BSc degree from
the University of Essex, England, in 1980, and
the MSc and PhD degrees from Rutgers
University, New Brunswick, New Jersey, in
1986 and 1990, respectively, all in electrical
engineering. He currently serves as a tenured
full professor in the Department of Electrical and
Computer Engineering at Carnegie Mellon Uni-
versity (CMU). Before joining CMU in August
2000, he was with the Electrical and Computer
Engineering Department at the State University of New York at Buffalo
(SUNY/Buffalo). He joined SUNY/Buffalo in 1990 as an assistant
professor, where he was granted early tenure and promoted to
associate professor in 1995, and to full professor in 1998. Prior to
joining academia, he was with Bell Communications Research (Bellcore)
between 1988-1990 doing research in optical networks and commu-
nication systems. His current research interests are in wireless networks
and communication systems, high-speed networking, optical networks,
satellite communications, bioinformatics, and security. He has published
more than 200 technical papers in IEEE journals and conference
proceedings. He is well-known for his contributions in optical networks
(optical transmission-access-networking, especially radio over fiber
networks, coherent lightwave transmission systems, and amplified
direct-detection lightwave transmission sytems) and his recent work
on iCAR (the Integrated Cellular and Ad Hoc Relay Systems) is
internationally acclaimed as well. He is the author of the book Ad Hoc
Wireless Networks: A Communication-Theoretic Perspective (Wiley and
Sons, 2006) and was also the architect of the High Performance
Waveform (HPW) that was implemented in Harris RF Communications
AN/PRC-117f UHF band man-pack tactical radio. His industrial
experience includes periods with Bell Communications Research, CTI
Inc., Harris RF Communications, Aria Wireless Systems, Clearwire
Technologies, Nokia Networks, Nokia Research Center, Neuro Kinetics,
Asea Brown Boveri (ABB), General Motors (GM), and Intel. He currently
serves as a consultant for several companies, law firms, and
government agencies in the US and Europe in the broad area of
telecommunications and networking. He is also a codirector (thrust
leader) of the Center for Wireless and Broadband Networking Research
at Carnegie Mellon University. More details about his research interests,
research group, projects, and publications can be found at http://
www.ece.cmu.edu/~tonguz/. In addition to serving on the technical
program committees of several IEEE conferences (such as INFOCOM,
GLOBECOM, ICC, VTC, and WCNC) and symposia in the areas of
wireless communications and optical networks, Dr. Tonguz currently
serves or has served as an associate editor for the IEEE Transactions
on Communications, IEEE Communications Magazine, and IEEE
Journal of Lightwave Technology. He was a guest editor of the special
issue of the IEEE Journal of Lightwave Technology and IEEE Journal on
Selected Areas in Communications on multiwavelength optical networks
and technology, published in 1996, and a guest editor of the special
issue of the Journal of Mobile Multimedia on advanced mobile
technologies for health care applications, published in 2006. He is a
member of the IEEE.
> For more information on this or any other computing topic,
please visit our Digital Library at www.computer.org/publications/dlib.

Optimal Transmit Power in Wireless Sensor Networks - 2006

Загружено:

Сведения о документе

Исходное описание:

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Optimal Transmit Power in Wireless Sensor Networks - 2006

Загружено:

Авторское право:

Доступные форматы

Optimal Transmit Power in

Wireless Sensor Networks

Вам также может понравиться